Crypto Pairs Trading (ETH/BTC)
Trade the mean-reverting spread between two cointegrated assets — here ETH vs BTC.
Failed honestly on eight years of real data — the pair was not reliably cointegrated, a random sign-flip beat the strategy, and one regime break ate the account.
- Why it fails
- The spread half-life was far too long to trade, the pair cleared no cointegration test, and the 2021 regime break drove a −93% drawdown on just 46 trades in eight years.
- When / how it stopped
- Tested over 2018–2026 on ETH/BTC. PF 0.69, Sharpe 0.10, −93.3% drawdown, 46 trades, placebo FAIL — 58% of random-sign permutations beat the real strategy.
Pairs trading is the canonical statistical-arbitrage idea: find two assets whose prices move together, trade the spread when it diverges, and collect as it reverts to the mean. The academic pedigree is real — Gatev, Goetzmann & Rouwenhorst (2006) documented profits on US equities — and ETH versus BTC is the obvious crypto candidate.
We ran a textbook cointegration version on ETH/BTC over 2018–2026 — z-score the spread, fade it at ±2σ, exit at 0, stop at ±3.5σ — through all 11 gates. It failed:
- Profit factor 0.69, Sharpe 0.10, max drawdown −93.3%, on just 46 trades in eight years ($5k → $2,349). The decisive result was the placebo: 58.3% of random-sign permutations beat the real strategy, whose PF sat well below the placebo 95th percentile (1.18).
- The pair was never cointegrated to begin with. Engle–Granger gives p = 0.32 and p = 0.65; the residual ADF p-value is 0.14; Johansen finds no cointegrating vector. The spread’s half-life of mean reversion was ~486 days — far too slow to trade — and the rolling cointegration fraction was just 7.6%.
- One year ate the account. The 2021 regime break cost −$6,116 as ETH repriced against BTC and the spread did not revert, driving the −93% drawdown.
This is consistent with the literature, not a refutation of it: pairs trading needs a relationship that stays cointegrated, and ETH/BTC simply did not over this window.
The full pre-registered report, gate scorecard and charts are in our validation write-up.
→ Read our full validation report: /strategy/pairs-eth-btc
Sources
- Our validation report — Pairs Trading ETH/BTC, placebo failed, not cointegrated
- Gatev, Goetzmann & Rouwenhorst (2006), 'Pairs Trading: Performance of a Relative-Value Arbitrage Rule', Review of Financial Studies
Frequently asked
Does crypto pairs trading work in 2026?
Not tradeably on ETH/BTC over 2018–2026. A z-score cointegration pairs strategy returned a profit factor of 0.69 with a −93.3% drawdown and a Sharpe of 0.10, and it failed the placebo: 58% of random-sign permutations beat the real strategy. The pair was not reliably cointegrated and the system took only 46 trades in eight years.
Why did the ETH/BTC pairs strategy have a 93% drawdown?
The 2021 regime break. ETH structurally repriced against BTC during 2021, and a mean-reversion spread trade — which assumes the ratio snaps back to its historical band — kept fading a move that did not revert. That single year was the dominant loss in the whole backtest. Cointegration measured on history is no guarantee the spread stays stationary, and here it did not.
Not investment advice — your mileage may vary, but the burden of proof is on the person claiming an edge. This entry describes general research and published evidence (or its absence), not a recommendation. See the full disclaimer.